Hacker News new | ask | show | jobs
by thesz 770 days ago
Note that KANs use LBFGS, which is second-order optimization method. My experience with the use of second-order features suggests that simple gradient descent often leads to divergence.